基于多尺度小波包分析的肺音特征提取与分类  

The Feature Extraction and Classification of Lung Sounds Based on Wavelet Packet Multiscale Analysis

作　　者：刘毅[1] 张彩明[1] 赵玉华[2] 董亮[3] 

机构地区：[1]山东大学计算机科学与技术学院,济南250061 [2]山东大学信息科学与工程学院,250100 [3]山东大学齐鲁医院呼吸内科,济南250012

出　　处：《计算机学报》2006年第5期769-777,共9页Chinese Journal of Computers

基　　金：山东省自然科学基金(Y2005G01)资助.

摘　　要：提出了一种适于非平稳肺音信号的特征提取方法.以4种肺音信号(正常、气管炎、肺炎和哮喘)为样本数据,通过分析肺音信号的时频分布特点,选择了具有任意多分辨分解特性的小波包.对小波包进行空间划分后找到了适合肺音特征提取的最优基,并基于最优基对肺音信号进行快速多尺度的分解,得到了各级节点的高维小波系数矩阵,建立了小波系数与信号能量在时域上的等价关系,并将能量作为特征值,构造了低维的作为分类神经网络的输入特征矢量,大大降低了输入特征的维数.研究表明该算法的识别性能是高效的.In this paper, a novel method of feature extraction in non-stable lung sound signals is put forward. Four kinds of lung sounds data（collected in the state of normal, bronchus, pneumonia and asthma respectively） are sampled from various subjects. By studying the time frequency distribution characteristics of the respiratory signals, the authors select the wavelet packets that have the trait of arbitrary distinction and decomposition. After space partition of wavelet packets, the best wavelet packet basis for feature extraction is picked out. Based on the best basis, we can do fast arbitrary muhi-scale WPT, and obtain each higher dimension wavelet coefficients matrix. And then the equabvalue relation in time domain between wavelet coefficients and signal energy is found. The energy is used as eigenvalue, and feature vectors of artificial neural network（ANN） for classification are formes. This greatly decreases the number of input vectors of ANN. Extensive experimental results demonstrate that the proposed feature extraction method has encouraging recognition performance.

关 键 词：肺音 多尺度分析 小波包 特征提取 分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]